Patrice Vignola
Software Engineer at Meta
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolPatrice Vignola is a software engineer with 11 years of experience specializing in ML infrastructure and GPU acceleration, currently enabling GenAI workloads on MTIA accelerators at Meta. Previously at Microsoft he led performance optimizations for ONNX Runtime and DirectML—pushing DirectX-based ML inference and training to rival CUDA on models like Stable Diffusion and large language models. He has deep low-level experience adding operators and kernels (DirectMLX, DirectML, tensorflow-directml) and improving runtime performance, memory management, and dynamic graph compilation. An active open-source contributor, Patrice has touched high-profile repos such as microsoft/DirectML and onnxruntime and even contributed to Ethereum clients and DirectX headers, reflecting a breadth across ML, graphics, and blockchain tooling. Based in Seattle, he combines systems-level C++/HLSL work with mentoring experience and a track record of shipping production-grade accelerator integration.
11 years of coding experience
7 years of employment as a software developer
Bachelor of Engineering - BE Software Engineering, Bachelor of Engineering - BE Software Engineering at École de technologie supérieure
English
Github Skills (26)
c-language10
operation10
directml10
tensorrt10
machine-learning10
go10
tensorflow10
onnxruntime10
wsl210
performance-optimization10
ethereum10
rest-api10
tensor10
cprogramming-language10
directx10
Programming languages (14)
Github contributions (5)
microsoft/onnxruntime
Oct 2019 - Jan 2023
ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator
Role in this project:
ML Engineer
ML EngineerContributions:1 release, 280 reviews, 71 commits in 3 years 3 months
Contributions summary:Patrice primarily focused on enhancing the DML (DirectML) backend for the ONNX Runtime, specifically addressing performance and functional gaps for machine learning inference. Their contributions included adding support for empty tensors, registering various data types for operators like "Where," and implementing the "FastGelu" operation. The user was also involved in fixing bugs, such as crashes within the "FusedMatMul" and "Attention" operators. Furthermore, the user contributed to performance improvements, such as enabling NCHW transpose, improved memory management, and dynamic graph compilation.
runtimetrainingtensorflowai-frameworkaccelerator
microsoft/DirectML
Jun 2020 - Aug 2022
DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.
Role in this project:ML Engineer
ML EngineerContributions:20 reviews, 24 commits, 40 PRs in 2 years 2 months
Contributions summary:Patrice primarily contributed to the `DirectMLX` library, adding new operators and functionalities. They implemented operators like `ScatterElements`, `ATanYX`, `SliceGrad`, `CumulativeSummation`, `CumulativeProduct`, `GatherND`, `TopK`, `Add`, `RoiAlign`, and `RoiAlignGrad`. Additionally, the user modified a SqueezeNet model script to use GPU instead of DML for device identification and also added an API for `DML_OPERATOR_ELEMENT_WISE_NEGATE`. The majority of the changes are directly related to the DirectX 12 library for machine learning.
amdaccelerationdirectmlgpu-accelerationnvidia
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.